kaggle竞赛的小结基于--House Prices: Advanced Regression Techniques（回归类问题）
前天报名了天池的天池精准医疗大赛——人工智能辅助糖尿病遗传风险预测的比赛，今天的提交排名是180多名，但是总排名300+，还没有进行数据工程的工作，所以去kaggle找了这个有相似度的题。下面将我这两天看的一些特征工程方面的代码贴在下面，数据的预处理可以按照这个步骤，其中kaggle的链接kaggle的处理参考
#!/usr/bin/python
#coding:utf-8
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
from scipy.stats import norm
from sklearn.preprocessing import StandardScaler
from scipy import stats
import warnings
import xgboost as xgb
warnings.filterwarnings('ignore')
df_train = pd.read_csv('data/train.csv')
'''
============================================================
0.总体属性的概览
============================================================
'''
# 显示数据的所有列元素
# print (df_train.columns)
'''
# 显示数据
print df_train['SalePrice'].describe()
sns.distplot(df_train['SalePrice'])
plt.show(sns.distplot(df_train['SalePrice']))
'''
'''
#skewness偏度 and kurtosis峰度
print("Skewness: %f" % df_train['SalePrice'].skew())
print("Kurtosis: %f" % df_train['SalePrice'].kurt())
'''
'''
============================================================
1.单一属性的分析
============================================================
'''
'''
#scatter plot grlivarea/saleprice 查看GrLivArea属性和最终预测值之间的可能关系
var = 'GrLivArea'
data = pd.concat([df_train['SalePrice'], df_train[var]], axis=1)
plt.show(data.plot.scatter(x=var, y='SalePrice', ylim=(0,800000)))
'''
'''
#scatter plot totalbsmtsf/saleprice 查看TotalBsmtSF属性和最终预测值之间的可能关系
var = 'TotalBsmtSF'
data = pd.concat([df_train['SalePrice'], df_train[var]], axis=1)
plt.show(data.plot.scatter(x=var, y='SalePrice', ylim=(0,800000)))
'''
'''
#box plot overallqual/saleprice 查看OverallQual属性和最终预测值之间的可能关系
var = 'OverallQual'
data = pd.concat([df_train['SalePrice'], df_train[var]], axis=1)
f, ax = plt.subplots(figsize=(8, 6))
fig = sns.boxplot(x=var, y="SalePrice", data=data)
plt.show(fig.axis(ymin=0, ymax=800000))
'''
'''
var = 'YearBuilt' # 查看YearBuilt年份和最终预测值之间的关系
data = pd.concat([df_train['SalePrice'], df_train[var]], axis=1)
f, ax = plt.subplots(figsize=(18, 8))
fig = sns.boxplot(x=var, y="SalePrice", data=data)
fig.axis(ymin=0, ymax=800000)
plt.show(plt.xticks(rotation=90))
# 用scatter表示
# plt.scatter(df_train['YearBuilt'], df_train['SalePrice'])
# plt.show()
'''
'''
============================================================
2.相关关系的分析
============================================================
'''
'''
#correlation matrix 所有属性的相关关系分析
corrmat = df_train.corr()
f, ax = plt.subplots(figsize=(18, 9))
plt.show(sns.heatmap(corrmat, vmax=.8, square=True))
'''
'''
#saleprice correlation matrix 最终预测值关联的相关矩阵，还有特征值之间的相关关系
corrmat = df_train.corr()
k = 11 #number of variables for heatmap
cols = corrmat.nlargest(k, 'SalePrice')['SalePrice'].index
cm = np.corrcoef(df_train[cols].values.T)
sns.set(font_scale=0.75)
hm = sns.heatmap(cm, cbar=True, annot=True, square=True, fmt='.2f', annot_kws={'size': 10}, yticklabels=cols.values, xticklabels=cols.values)
plt.show()
'''
'''
#scatterplot 关联特征值的其他显示形式，方便查看特征值之间的关联
sns.set()
cols = ['SalePrice', 'OverallQual', 'GrLivArea', 'GarageCars', 'TotalBsmtSF', 'FullBath', 'YearBuilt']
sns.pairplot(df_train[cols], size=2.5)
plt.show()
'''
'''
============================================================
3.缺失值的处理
============================================================
'''
'''
#missing data 缺失值的查询和显示
# 当缺失值大于15%时，直接删掉数据
total = df_train.isnull().sum().sort_values(ascending=False)
percent = (df_train.isnull().sum()/df_train.isnull().count()).sort_values(ascending=False)
missing_data = pd.concat([total, percent], axis=1, keys=['Total', 'Percent'])
print missing_data.head(20)
#dealing with missing data
df_train = df_train.drop((missing_data[missing_data['Total'] > 1]).index,1) # 删除上述前18个特征
df_train = df_train.drop(df_train.loc[df_train['Electrical'].isnull()].index) # 删除 Electrical 取值丢失的样本
print "isnull", df_train.isnull().sum().max() #just checking that there's no missing data missing...
'''
'''
============================================================
4.异常点的分析
============================================================
'''
'''
#standardizing data 将预测值的数据标准化(0,1),检测出较大的异常那些数据(单一变量的异常)
saleprice_scaled = StandardScaler().fit_transform(df_train['SalePrice'][:, np.newaxis])
low_range = saleprice_scaled[saleprice_scaled[:, 0].argsort()][:10]
high_range = saleprice_scaled[saleprice_scaled[:, 0].argsort()][-10:]
print('outer range (low) of the distribution:')
print(low_range)
print('\nouter range (high) of the distribution:')
print(high_range)
'''
'''
#bivariate analysis saleprice/grlivarea相关属性异常点的检测(两个变量的异常)
var = 'GrLivArea'
data = pd.concat([df_train['SalePrice'], df_train[var]], axis=1)
data.plot.scatter(x=var, y='SalePrice', ylim=(0, 800000))
plt.show()
# 删除两个异常点
df_train.sort_values(by = 'GrLivArea', ascending = False)[:2]
df_train = df_train.drop(df_train[df_train['Id'] == 1299].index)
df_train = df_train.drop(df_train[df_train['Id'] == 524].index)
'''
'''
#histogram and normal probability plot 查看预测的属性是否标准分布，标准
# 的正态分布对结果具有很好的效果
sns.distplot(df_train['SalePrice'], fit=norm);
fig = plt.figure()
res = stats.probplot(df_train['SalePrice'], plot=plt)
plt.show()
'''
'''
#applying log transformation 将预测属性标准正态化
df_train['SalePrice'] = np.log(df_train['SalePrice'])
#transformed histogram and normal probability plot
sns.distplot(df_train['SalePrice'], fit=norm);
fig = plt.figure()
res = stats.probplot(df_train['SalePrice'], plot=plt)
plt.show()
'''
'''
#histogram and normal probability plot 查看GrLivArea属性是否符合正态分布
sns.distplot(df_train['GrLivArea'], fit=norm)
fig = plt.figure()
res = stats.probplot(df_train['GrLivArea'], plot=plt)
plt.show()
'''
'''
#data transformation 将GrLivArea属性转换成符合正态分布
df_train['GrLivArea'] = np.log(df_train['GrLivArea'])
sns.distplot(df_train['GrLivArea'], fit=norm)
fig = plt.figure()
res = stats.probplot(df_train['GrLivArea'], plot=plt)
plt.show()
'''
这里，在参加比赛几天，贴上一个stacking的代码，特征工程和参数都没有怎么做，但是这是一个stacking的很好的例子，含有CV验证，只需要将自己的特征工程拷贝到本程序的相应位置。
#coding: UTF-8
import numpy as np
import pandas as pd
from dateutil.parser import parse
from sklearn.cross_validation import KFold
from scipy.stats import norm, skew
from scipy.special import boxcox1p
data_path = 'data/'
train = pd.read_csv(data_path+'d_train_20180102.csv')
test = pd.read_csv(data_path+'d_test_A_20180102.csv')
train['血糖'] = np.log(train['血糖'])
y_train = train['血糖'].values
ntrain = train.shape[0]
ntest = test.shape[0]
def make_feat(train,test):
train_id = train.id.values.copy()
test_id = test.id.values.copy()
data = pd.concat([train, test])
data = data.drop(['血糖'], axis=1)
data['性别'] = data['性别'].map({'男':1,'女':0,None: -1})
data['体检日期'] = (pd.to_datetime(data['体检日期']) - parse('2016-10-09')).dt.days
data['*r-谷氨酰基转换酶'] = data['*r-谷氨酰基转换酶'].fillna(data['*r-谷氨酰基转换酶'].median())
#data['*r-谷氨酰基转换酶'] = np.log1p(data['*r-谷氨酰基转换酶'])
data['*丙氨酸氨基转换酶'] = data['*丙氨酸氨基转换酶'].fillna(data['*丙氨酸氨基转换酶'].median())
#data['*丙氨酸氨基转换酶'] = np.log1p(data['*丙氨酸氨基转换酶'])
data['*天门冬氨酸氨基转换酶'] = data['*天门冬氨酸氨基转换酶'].fillna(data['*天门冬氨酸氨基转换酶'].median())
#data['*天门冬氨酸氨基转换酶'] = np.log1p(data['*天门冬氨酸氨基转换酶'])
data['*总蛋白'] = data['*总蛋白'].fillna(data['*总蛋白'].median())
#data['*总蛋白'] = np.log1p(data['*总蛋白'])
data['*球蛋白'] = data['*球蛋白'].fillna(data['*球蛋白'].median())
#data['*球蛋白'] = np.log1p(data['*球蛋白'])
data['*碱性磷酸酶'] = data['*碱性磷酸酶'].fillna(data['*碱性磷酸酶'].median())
#data['*碱性磷酸酶'] = np.log1p(data['*碱性磷酸酶'])
data['中性粒细胞%'] = data['中性粒细胞%'].fillna(data['中性粒细胞%'].median())
#data['中性粒细胞%'] = np.log1p(data['中性粒细胞%'])
data['乙肝e抗体'] = data['乙肝e抗体'].fillna(data['乙肝e抗体'].median())
#data['乙肝e抗体'] = np.log1p(data['乙肝e抗体'])
data['乙肝e抗原'] = data['乙肝e抗原'].fillna(data['乙肝e抗原'].median())
#data['乙肝e抗原'] = np.log1p(data['乙肝e抗原'])
data['乙肝核心抗体'] = data['乙肝核心抗体'].fillna(data['乙肝核心抗体'].median())
#data['乙肝核心抗体'] = np.log1p(data['乙肝核心抗体'])
data['乙肝表面抗体'] = data['乙肝表面抗体'].fillna(data['乙肝表面抗体'].median())
#data['乙肝表面抗体'] = np.log1p(data['乙肝表面抗体'])
data['乙肝表面抗原'] = data['乙肝表面抗原'].fillna(data['乙肝表面抗原'].median())
#data['乙肝表面抗原'] = np.log1p(data['乙肝表面抗原'])
data['单核细胞%'] = data['单核细胞%'].fillna(data['单核细胞%'].median())
#data['单核细胞%'] = np.log1p(data['单核细胞%'])
data['嗜碱细胞%'] = data['嗜碱细胞%'].fillna(data['嗜碱细胞%'].median())
#data['嗜碱细胞%'] = np.log1p(data['嗜碱细胞%'])
data['嗜酸细胞%'] = data['嗜酸细胞%'].fillna(data['嗜酸细胞%'].median())
#data['嗜酸细胞%'] = np.log1p(data['嗜酸细胞%'])
data['尿素'] = data['尿素'].fillna(data['尿素'].median())
#data['尿素'] = np.log1p(data['尿素'])
data['尿酸'] = data['尿酸'].fillna(data['尿酸'].median())
#data['尿酸'] = np.log1p(data['尿酸'])
data['总胆固醇'] = data['总胆固醇'].fillna(data['总胆固醇'].median())
#data['总胆固醇'] = np.log1p(data['总胆固醇'])
data['淋巴细胞%'] = data['淋巴细胞%'].fillna(data['淋巴细胞%'].median())
#data['淋巴细胞%'] = np.log1p(data['淋巴细胞%'])
data['甘油三酯'] = data['甘油三酯'].fillna(data['甘油三酯'].median())
#data['甘油三酯'] = np.log1p(data['甘油三酯'])
data['白球比例'] = data['白球比例'].fillna(data['白球比例'].median())
#data['白球比例'] = np.log1p(data['白球比例'])
data['白细胞计数'] = data['白细胞计数'].fillna(data['白细胞计数'].median())
#data['白细胞计数'] = np.log1p(data['白细胞计数'])
data['白蛋白'] = data['白蛋白'].fillna(data['白蛋白'].median())
#data['白蛋白'] = np.log1p(data['白蛋白'])
data['红细胞体积分布宽度'] = data['红细胞体积分布宽度'].fillna(data['红细胞体积分布宽度'].median())
#data['红细胞体积分布宽度'] = np.log1p(data['红细胞体积分布宽度'])
data['红细胞压积'] = data['红细胞压积'].fillna(data['红细胞压积'].median())
#data['红细胞压积'] = np.log1p(data['红细胞压积'])
data['红细胞平均体积'] = data['红细胞平均体积'].fillna(data['红细胞平均体积'].median())
#data['红细胞平均体积'] = np.log1p(data['红细胞平均体积'])
data['红细胞平均血红蛋白浓度'] = data['红细胞平均血红蛋白浓度'].fillna(data['红细胞平均血红蛋白浓度'].median())
#data['红细胞平均血红蛋白浓度'] = np.log1p(data['红细胞平均血红蛋白浓度'])
data['红细胞平均血红蛋白量'] = data['红细胞平均血红蛋白量'].fillna(data['红细胞平均血红蛋白量'].median())
#data['红细胞平均血红蛋白量'] = np.log1p(data['红细胞平均血红蛋白量'])
data['红细胞计数'] = data['红细胞计数'].fillna(data['红细胞计数'].median())
#data['红细胞计数'] = np.log1p(data['红细胞计数'])
data['肌酐'] = data['肌酐'].fillna(data['肌酐'].median())
#data['肌酐'] = np.log1p(data['肌酐'])
data['血小板体积分布宽度'] = data['血小板体积分布宽度'].fillna(data['血小板体积分布宽度'].median())
#data['血小板体积分布宽度'] = np.log1p(data['血小板体积分布宽度'])
data['血小板平均体积'] = data['血小板平均体积'].fillna(data['血小板平均体积'].median())
#data['血小板平均体积'] = np.log1p(data['血小板平均体积'])
data['血小板比积'] = data['血小板比积'].fillna(data['血小板比积'].median())
#data['血小板比积'] = np.log1p(data['血小板比积'])
data['血小板计数'] = data['血小板计数'].fillna(data['血小板计数'].median())
#data['血小板计数'] = np.log1p(data['血小板计数'])
data['血红蛋白'] = data['血红蛋白'].fillna(data['血红蛋白'].median())
#data['血红蛋白'] = np.log1p(data['血红蛋白'])
data['高密度脂蛋白胆固醇'] = data['高密度脂蛋白胆固醇'].fillna(data['高密度脂蛋白胆固醇'].median())
#data['高密度脂蛋白胆固醇'] = np.log1p(data['高密度脂蛋白胆固醇'])
data['低密度脂蛋白胆固醇'] = data['低密度脂蛋白胆固醇'].fillna(data['低密度脂蛋白胆固醇'].median())
#data['低密度脂蛋白胆固醇'] = np.log1p(data['低密度脂蛋白胆固醇'])
data['性别'] = data['性别'].fillna(1)
data.fillna(data.median(axis=0))
'''
numeric_features = data.dtypes[data.dtypes != "object"].index
skewed = data[numeric_features].apply(lambda x: skew(x.dropna().astype(float)))
skewed = skewed[abs(skewed) > 0.75]
skewed = skewed.index
data[skewed] = np.log1p(data[skewed])
'''
train_feat = np.array(data[:train.shape[0]])
test_feat = np.array(data[train.shape[0]:])
print "train_feat======================", train_feat.shape
'''
total = data.isnull().sum().sort_values(ascending=False)
percent = (data.isnull().sum()/data.isnull().count()).sort_values(ascending=False)
missing_data = pd.concat([total, percent], axis=1, keys=['Total', 'Percent'])
print missing_data.head(20)
'''
return train_feat, test_feat
x_train, x_test = make_feat(train, test)
########################################################################
import xgboost as xgb
from sklearn.cross_validation import KFold
from sklearn.ensemble import ExtraTreesRegressor
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error
from sklearn.linear_model import Ridge, RidgeCV, ElasticNet, LassoCV, Lasso
from math import sqrt
NFOLDS = 10
SEED = 5
SUBMISSION_FILE = 'sample_submission.csv'
kf = KFold(ntrain, n_folds=NFOLDS, shuffle=True, random_state=SEED)
class SklearnWrapper(object):
def __init__(self, clf, seed=0, params=None):
params['random_state'] = seed
self.clf = clf(**params)
def train(self, x_train, y_train):
self.clf.fit(x_train, y_train)
def predict(self, x):
return self.clf.predict(x)
class XgbWrapper(object):
def __init__(self, seed=0, params=None):
self.param = params
self.param['seed'] = seed
self.nrounds = params.pop('nrounds', 250)
def train(self, x_train, y_train):
dtrain = xgb.DMatrix(x_train, label=y_train)
self.gbdt = xgb.train(self.param, dtrain, self.nrounds)
def predict(self, x):
return self.gbdt.predict(xgb.DMatrix(x))
def get_oof(clf):
oof_train = np.zeros((ntrain,))
oof_test = np.zeros((ntest,))
oof_test_skf = np.empty((NFOLDS, ntest))
for i, (train_index, test_index) in enumerate(kf):
x_tr = x_train[train_index]
y_tr = y_train[train_index]
x_te = x_train[test_index]
clf.train(x_tr, y_tr)
oof_train[test_index] = clf.predict(x_te)
oof_test_skf[i, :] = clf.predict(x_test)
oof_test[:] = oof_test_skf.mean(axis=0)
return oof_train.reshape(-1, 1), oof_test.reshape(-1, 1)
et_params = {
'n_jobs': 16,
'n_estimators': 100,
'max_features': 0.5,
'max_depth': 12,
'min_samples_leaf': 2,
}
rf_params = {
'n_jobs': 16,
'n_estimators': 100,
'max_features': 0.2,
'max_depth': 12,
'min_samples_leaf': 2,
}
xgb_params = {
'seed': 0,
'colsample_bytree': 0.7,
'silent': 1,
'subsample': 0.7,
'learning_rate': 0.075,
'objective': 'reg:linear',
'max_depth': 4,
'num_parallel_tree': 1,
'min_child_weight': 1,
'eval_metric': 'rmse',
'nrounds': 500
}
rd_params={
'alpha': 10
}
ls_params={
'alpha': 0.005
}
xg = XgbWrapper(seed=SEED, params=xgb_params)
et = SklearnWrapper(clf=ExtraTreesRegressor, seed=SEED, params=et_params)
rf = SklearnWrapper(clf=RandomForestRegressor, seed=SEED, params=rf_params)
rd = SklearnWrapper(clf=Ridge, seed=SEED, params=rd_params)
ls = SklearnWrapper(clf=Lasso, seed=SEED, params=ls_params)
xg_oof_train, xg_oof_test = get_oof(xg)
et_oof_train, et_oof_test = get_oof(et)
rf_oof_train, rf_oof_test = get_oof(rf)
rd_oof_train, rd_oof_test = get_oof(rd)
ls_oof_train, ls_oof_test = get_oof(ls)
print("XG-CV: {}".format(0.5*(mean_squared_error(np.exp(y_train), np.exp(xg_oof_train)))))
print("ET-CV: {}".format(0.5*(mean_squared_error(np.exp(y_train), np.exp(et_oof_train)))))
print("RF-CV: {}".format(0.5*(mean_squared_error(np.exp(y_train), np.exp(rf_oof_train)))))
print("RD-CV: {}".format(0.5*(mean_squared_error(np.exp(y_train), np.exp(ls_oof_train)))))
x_train = np.concatenate((xg_oof_train, et_oof_train, rf_oof_train, rd_oof_train, ls_oof_train), axis=1)
x_test = np.concatenate((xg_oof_test, et_oof_test, rf_oof_test, rd_oof_test, ls_oof_test), axis=1)
print("{},{}".format(x_train.shape, x_test.shape))
dtrain = xgb.DMatrix(x_train, label=y_train)
dtest = xgb.DMatrix(x_test)
xgb_params = {
'seed': 0,
'colsample_bytree': 0.8,
'silent': 1,
'subsample': 0.6,
'learning_rate': 0.01,
'objective': 'reg:linear',
'max_depth': 1,
'num_parallel_tree': 1,
'min_child_weight': 1,
'eval_metric': 'rmse',
}
res = xgb.cv(xgb_params, dtrain, num_boost_round=1000, nfold=4, seed=SEED, stratified=False,
early_stopping_rounds=25, verbose_eval=10, show_stdv=True)
best_nrounds = res.shape[0] - 1
cv_mean = res.iloc[-1, 0]
cv_std = res.iloc[-1, 1]
print('Ensemble-CV: {0}+{1}'.format(cv_mean, cv_std))
gbdt = xgb.train(xgb_params, dtrain, best_nrounds)
'''
submission = pd.read_csv(SUBMISSION_FILE)
submission.iloc[:, 1] = gbdt.predict(dtest)
saleprice = np.exp(submission['血糖'])
submission['SalePrice'] = saleprice
submission.to_csv('xgstacker_starter.sub.csv', index=None)
'''
y_pred = np.exp(gbdt.predict(dtest))
xgb_pred_df = pd.DataFrame(y_pred)
xgb_pred_df = xgb_pred_df.round(3)
xgb_pred_df.to_csv('xgstacker.csv', header=None, index=False, float_format="%.4f")
在这里，我在贴一个ensemble的代码，使用的时候必须修改其中的RMSE，来保证cv的正确公式，得到比较好的本地得分，这里记录一下LB和CV可能得到不一样的分数，这和x的分布改变有关系。
#coding: UTF-8
import numpy as np
import pandas as pd
from dateutil.parser import parse
from sklearn.cross_validation import KFold
from scipy.stats import norm, skew
from scipy.special import boxcox1p
from sklearn.preprocessing import StandardScaler, RobustScaler
data_path = 'data/'
train = pd.read_csv(data_path+'d_train_20180102.csv')
test = pd.read_csv(data_path+'d_test_A_20180102.csv')
def make_feat(train,test):
train_id = train.id.values.copy()
test_id = test.id.values.copy()
data = pd.concat([train, test])
data = data.drop(['血糖'], axis=1)
data['性别'] = data['性别'].map({'男':1,'女':0,None: -1})
data['体检日期'] = (pd.to_datetime(data['体检日期']) - parse('2016-10-09')).dt.days
data['*r-谷氨酰基转换酶'] = data['*r-谷氨酰基转换酶'].fillna(data['*r-谷氨酰基转换酶'].median())
#data['*r-谷氨酰基转换酶'] = np.log1p(data['*r-谷氨酰基转换酶'])
data['*丙氨酸氨基转换酶'] = data['*丙氨酸氨基转换酶'].fillna(data['*丙氨酸氨基转换酶'].median())
#data['*丙氨酸氨基转换酶'] = np.log1p(data['*丙氨酸氨基转换酶'])
data['*天门冬氨酸氨基转换酶'] = data['*天门冬氨酸氨基转换酶'].fillna(data['*天门冬氨酸氨基转换酶'].median())
#data['*天门冬氨酸氨基转换酶'] = np.log1p(data['*天门冬氨酸氨基转换酶'])
data['*总蛋白'] = data['*总蛋白'].fillna(data['*总蛋白'].median())
#data['*总蛋白'] = np.log1p(data['*总蛋白'])
data['*球蛋白'] = data['*球蛋白'].fillna(data['*球蛋白'].median())
#data['*球蛋白'] = np.log1p(data['*球蛋白'])
data['*碱性磷酸酶'] = data['*碱性磷酸酶'].fillna(data['*碱性磷酸酶'].median())
#data['*碱性磷酸酶'] = np.log1p(data['*碱性磷酸酶'])
data['中性粒细胞%'] = data['中性粒细胞%'].fillna(data['中性粒细胞%'].median())
#data['中性粒细胞%'] = np.log1p(data['中性粒细胞%'])
data['乙肝e抗体'] = data['乙肝e抗体'].fillna(data['乙肝e抗体'].median())
#data['乙肝e抗体'] = np.log1p(data['乙肝e抗体'])
data['乙肝e抗原'] = data['乙肝e抗原'].fillna(data['乙肝e抗原'].median())
#data['乙肝e抗原'] = np.log1p(data['乙肝e抗原'])
data['乙肝核心抗体'] = data['乙肝核心抗体'].fillna(data['乙肝核心抗体'].median())
#data['乙肝核心抗体'] = np.log1p(data['乙肝核心抗体'])
data['乙肝表面抗体'] = data['乙肝表面抗体'].fillna(data['乙肝表面抗体'].median())
#data['乙肝表面抗体'] = np.log1p(data['乙肝表面抗体'])
data['乙肝表面抗原'] = data['乙肝表面抗原'].fillna(data['乙肝表面抗原'].median())
#data['乙肝表面抗原'] = np.log1p(data['乙肝表面抗原'])
data['单核细胞%'] = data['单核细胞%'].fillna(data['单核细胞%'].median())
#data['单核细胞%'] = np.log1p(data['单核细胞%'])
data['嗜碱细胞%'] = data['嗜碱细胞%'].fillna(data['嗜碱细胞%'].median())
#data['嗜碱细胞%'] = np.log1p(data['嗜碱细胞%'])
data['嗜酸细胞%'] = data['嗜酸细胞%'].fillna(data['嗜酸细胞%'].median())
#data['嗜酸细胞%'] = np.log1p(data['嗜酸细胞%'])
data['尿素'] = data['尿素'].fillna(data['尿素'].median())
#data['尿素'] = np.log1p(data['尿素'])
data['尿酸'] = data['尿酸'].fillna(data['尿酸'].median())
#data['尿酸'] = np.log1p(data['尿酸'])
data['总胆固醇'] = data['总胆固醇'].fillna(data['总胆固醇'].median())
#data['总胆固醇'] = np.log1p(data['总胆固醇'])
data['淋巴细胞%'] = data['淋巴细胞%'].fillna(data['淋巴细胞%'].median())
#data['淋巴细胞%'] = np.log1p(data['淋巴细胞%'])
data['甘油三酯'] = data['甘油三酯'].fillna(data['甘油三酯'].median())
#data['甘油三酯'] = np.log1p(data['甘油三酯'])
data['白球比例'] = data['白球比例'].fillna(data['白球比例'].median())
#data['白球比例'] = np.log1p(data['白球比例'])
data['白细胞计数'] = data['白细胞计数'].fillna(data['白细胞计数'].median())
#data['白细胞计数'] = np.log1p(data['白细胞计数'])
data['白蛋白'] = data['白蛋白'].fillna(data['白蛋白'].median())
#data['白蛋白'] = np.log1p(data['白蛋白'])
data['红细胞体积分布宽度'] = data['红细胞体积分布宽度'].fillna(data['红细胞体积分布宽度'].median())
#data['红细胞体积分布宽度'] = np.log1p(data['红细胞体积分布宽度'])
data['红细胞压积'] = data['红细胞压积'].fillna(data['红细胞压积'].median())
#data['红细胞压积'] = np.log1p(data['红细胞压积'])
data['红细胞平均体积'] = data['红细胞平均体积'].fillna(data['红细胞平均体积'].median())
#data['红细胞平均体积'] = np.log1p(data['红细胞平均体积'])
data['红细胞平均血红蛋白浓度'] = data['红细胞平均血红蛋白浓度'].fillna(data['红细胞平均血红蛋白浓度'].median())
#data['红细胞平均血红蛋白浓度'] = np.log1p(data['红细胞平均血红蛋白浓度'])
data['红细胞平均血红蛋白量'] = data['红细胞平均血红蛋白量'].fillna(data['红细胞平均血红蛋白量'].median())
#data['红细胞平均血红蛋白量'] = np.log1p(data['红细胞平均血红蛋白量'])
data['红细胞计数'] = data['红细胞计数'].fillna(data['红细胞计数'].median())
#data['红细胞计数'] = np.log1p(data['红细胞计数'])
data['肌酐'] = data['肌酐'].fillna(data['肌酐'].median())
#data['肌酐'] = np.log1p(data['肌酐'])
data['血小板体积分布宽度'] = data['血小板体积分布宽度'].fillna(data['血小板体积分布宽度'].median())
#data['血小板体积分布宽度'] = np.log1p(data['血小板体积分布宽度'])
data['血小板平均体积'] = data['血小板平均体积'].fillna(data['血小板平均体积'].median())
#data['血小板平均体积'] = np.log1p(data['血小板平均体积'])
data['血小板比积'] = data['血小板比积'].fillna(data['血小板比积'].median())
#data['血小板比积'] = np.log1p(data['血小板比积'])
data['血小板计数'] = data['血小板计数'].fillna(data['血小板计数'].median())
#data['血小板计数'] = np.log1p(data['血小板计数'])
data['血红蛋白'] = data['血红蛋白'].fillna(data['血红蛋白'].median())
#data['血红蛋白'] = np.log1p(data['血红蛋白'])
data['高密度脂蛋白胆固醇'] = data['高密度脂蛋白胆固醇'].fillna(data['高密度脂蛋白胆固醇'].median())
#data['高密度脂蛋白胆固醇'] = np.log1p(data['高密度脂蛋白胆固醇'])
data['低密度脂蛋白胆固醇'] = data['低密度脂蛋白胆固醇'].fillna(data['低密度脂蛋白胆固醇'].median())
#data['低密度脂蛋白胆固醇'] = np.log1p(data['低密度脂蛋白胆固醇'])
data['性别'] = data['性别'].fillna(1)
data.fillna(data.median(axis=0))
'''
numeric_features = data.dtypes[data.dtypes != "object"].index
skewed = data[numeric_features].apply(lambda x: skew(x.dropna().astype(float)))
skewed = skewed[abs(skewed) > 0.8]
skewed = skewed.index
data[skewed] = np.log1p(data[skewed])
scaler = StandardScaler()
scaler.fit(data[numeric_features])
scaled = scaler.transform(data[numeric_features])
for i, col in enumerate(numeric_features):
data[col] = scaled[:, i]
'''
train_feat = data[data.id.isin(train_id)]
test_feat = data[data.id.isin(test_id)]
print "train_feat======================", train_feat.shape
'''
total = data.isnull().sum().sort_values(ascending=False)
percent = (data.isnull().sum()/data.isnull().count()).sort_values(ascending=False)
missing_data = pd.concat([total, percent], axis=1, keys=['Total', 'Percent'])
print missing_data.head(20)
'''
return train_feat, test_feat
train_feat, test_feat = make_feat(train, test)
predictors = [f for f in test_feat.columns if f not in ['血糖']]
###################################################################
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import KFold
import xgboost as xgb
from sklearn.linear_model import Lasso, Ridge, ElasticNet
from sklearn.kernel_ridge import KernelRidge
from sklearn.base import BaseEstimator, RegressorMixin
from sklearn.preprocessing import LabelEncoder
from scipy.stats import skew
class CustomEnsembleRegressor(BaseEstimator, RegressorMixin):
def __init__(self, regressors=None):
self.regressors = regressors
def fit(self, X, y):
for regressor in self.regressors:
regressor.fit(X, y)
def predict(self, X):
self.predictions_ = list()
for regressor in self.regressors:
self.predictions_.append(np.exp(regressor.predict(X).ravel()))
return np.log1p(np.mean(self.predictions_, axis=0))
################################################################################
# RMSE
def rmse(y_true, y_pred):
return 0.5*(mean_squared_error(np.exp(y_true), np.exp(y_pred)))
# Cross-validation
def evaludate_model(model, x, y):
# print('Cross_validation..')
n_splits_val = 3
kf = KFold(n_splits=n_splits_val, shuffle=False)
idx = 0
rmse_buf = np.empty(n_splits_val)
for train, test in kf.split(x):
model.fit(x.iloc[train], y.iloc[train])
y_cv = model.predict(x.iloc[test])
rmse_buf[idx] = rmse(y.iloc[test], y_cv)
print('Interation #' + str(idx) + ': MSE = %.5f' % rmse_buf[idx])
idx += 1
mean_rmse = np.mean(rmse_buf)
print(' Mean MSE = %.5f' % mean_rmse + ' +/- %.5f' % np.std(rmse_buf))
return mean_rmse
def evaludate_submodels(models, x, y):
# print('Cross_validation..')
n_splits_val = 10
kf = KFold(n_splits=n_splits_val, shuffle=False)
for m_i, model in enumerate(models.regressors):
rmse_buf = np.empty(n_splits_val)
idx = 0
for train, test in kf.split(x):
model.fit(x.iloc[train], y.iloc[train])
y_cv = model.predict(x.iloc[test])
rmse_buf[idx] = rmse(y.iloc[test], y_cv)
# print('Interation #' + str(idx) + ': RMSE = %.5f' % rmse_buf[idx])
idx += 1
mean_rmse = np.mean(rmse_buf)
print('Model #' + str(m_i) + ': mean RMSE = %.5f' % mean_rmse + \
' +/- %.5f' % np.std(rmse_buf))
train_df_munged, test_df_munged = train_feat, test_feat
label_df = pd.DataFrame(index=train_df_munged.index, columns=['血糖'])
label_df['血糖'] = np.log(train['血糖'])
################################################################################
regr1 = xgb.XGBRegressor(
colsample_bytree=0.2,
gamma=0.0,
learning_rate=0.01,
max_depth=4,
min_child_weight=1.5,
n_estimators=30000,
reg_alpha=0.9,
reg_lambda=0.6,
subsample=0.2,
seed=42,
silent=1)
best_alpha = 0.00098
regr2 = Lasso(alpha=best_alpha, max_iter=50000)
regr3 = ElasticNet(alpha=0.001)
regr4 = KernelRidge(alpha=0.3, kernel='polynomial', degree=2, coef0=1.85)
regr = CustomEnsembleRegressor([regr1, regr2, regr3])
# Evaluation was commented to make it run as kernel
print('Evaluating each model separately..')
evaludate_submodels(regr, train_df_munged, label_df)
print('Evaluating ensemble..')
evaludate_model(regr, train_df_munged, label_df)
print('Fitting ensemble and predicting..')
# Fit the ensemble
regr.fit(train_df_munged, label_df)
# Run prediction on the Kaggle test set.
y_pred = regr.predict(test_df_munged)
################################################################################
print('Saving results..')
# Blend the results of the two regressors and save the prediction to a CSV file.
y_pred = np.exp(y_pred)
sub = pd.DataFrame()
sub['血糖'] = y_pred
sub.to_csv('ensemble_output.csv', header=False, index=False)
这些都是我在kaggle上面自己拔下来的代码，我觉得还是去kaggle上看开放的kernel比较好。